the cloud bill has gotten genuinely silly and most of my load is steady and predictable, not spiky at all. im seriously pricing out a couple of boxes in a colo again like the old days. i know all the arguments, ops burden, no instant scaling, hardware dies at 3am. but for steady baseline the math is lopsided. anyone actually done cloud to colo recently and regretted it, or didnt?
did it about two years ago for the baseline and kept a small cloud footprint for overflow and failover. zero regret on the money. the thing people underestimate isnt the hardware itself, its that you now own the whole networking, ipmi, and remote hands relationship. budget time for that, not just the box price.
we all ran colo before cloud even existed and the world didnt end. hardware dying at 3am is exactly what redundant boxes and a remote hands contract are for. the skills arent lost, theyre just unfashionable right now, and unfashionable is where the savings hide.
The egress savings alone made it worth it for me, which I was moaning about in another thread this week. Steady predictable traffic is precisely the profile where owning the pipe crushes per gb cloud pricing. Spiky early stage startups should stay in cloud, established operators with a flat baseline are quietly overpaying.
the remote hands relationship is the bit i keep forgetting to actually price in, youre right max. im not trying to go pure colo anyway, same as you, baseline on metal and cloud for the overflow and disaster recovery. that split feels like the real sweet spot rather than picking a side.